Systems-Level Modeling of Cancer Evolution

Abstract

Cancer is an evolutionary process: tumour cells acquire mutations, compete for resources, adapt to microenvironmental pressures, and evolve resistance to therapy through Darwinian selection. Understanding tumour evolution is essential for predicting treatment response, anticipating resistance mechanisms, and designing adaptive therapy strategies. Computational models of cancer evolution span multiple scales: population genetics models track allele frequencies across clonal populations, phylogenetic models reconstruct tumour evolutionary histories from sequencing data, agent-based models simulate individual cell dynamics within the tumour microenvironment, game-theoretic models analyse competitive interactions between tumour subclones, and machine learning models predict evolutionary trajectories from multi-omics data. We present the Cancer Evolution Modelling Framework (CEMF), evaluating five modelling approaches -- Wright-Fisher population dynamics, tumour phylogenetics from single-cell data, spatial agent-based evolutionary models, evolutionary game theory, and deep learning evolutionary predictors -- across four cancer evolution scenarios (clonal expansion in early tumourigenesis, therapy-driven selection, metastatic seeding dynamics, and adaptive therapy optimisation). Our Cancer Evolution Model Score (CEMS) measures evolutionary reconstruction accuracy, therapy resistance prediction, spatial dynamics capture, computational scalability, and clinical actionability. Deep learning evolutionary predictors achieve the highest CEMS (0.926) through integration of multi-timepoint sequencing data with learned evolutionary dynamics, while evolutionary game theory achieves the highest clinical actionability (0.960) through directly optimisable adaptive therapy schedules.